Skip to content
MAF Learning Hub
Advanced Duration: 25 min

Connectors and MCP tool binding

Prerequisites: deploying-agents-cloudhub

An agent is only as capable as the systems it can reach. This lesson connects two ideas you have already met - Anypoint Connectors and the Model Context Protocol - to give deployed agents safe, reusable access to enterprise systems.

What you will learn

  • What Anypoint Connectors provide and why to reuse them.
  • How a connector-backed capability becomes an MCP tool.
  • How binding keeps system access governed rather than ad hoc.

Anypoint Connectors

Anypoint Connectors are prebuilt, reusable integrations to systems like Salesforce, databases, SAP, and hundreds more. Rather than hand-coding access to each system, an agent reuses a connector-backed API - inheriting its reliability and management (MuleSoft documentation).

From connector to MCP tool

To let an agent use that capability, you expose it as an MCP tool (see Track 1’s MCP lesson) and bind it to the agent. The connector does the system work; MCP is the standardized way the agent invokes it.

# agent-network.yaml (excerpt)
mcpServers:
  - id: salesforce-tools
    url: https://mcp.example.com/salesforce   # backed by an Anypoint Connector
    auth: oauth2
agents:
  - id: account-agent
    tools:
      - mcpServer: salesforce-tools

Reuse, do not re-grant

Binding a connector-backed MCP tool means the agent gets exactly the governed capability you exposed - not raw credentials to the underlying system.

Keeping it governed

Because the bound tool is an MCP server behind the Omni Gateway, the MCP access policies from Track 3 (ABAC, rate limiting) still apply. Connectivity and governance stay coupled. (Forward-looking: exact connector-to-MCP packaging steps evolve; confirm against current official documentation.)

Next steps

Continue to Observe with Agent Visualizer and Anypoint Monitoring to watch these agents and tool calls in production.

References